With test-case gdb.server/unittest.exp and host board local-remote-host.exp I
run into:
...
builtin_spawn build/gdbserver/gdbserver --selftest^M
ERROR: : spawn id exp7 not open
while executing
"expect {
-i exp7 -timeout 10
-i $server_spawn_id
-re "Ran ($decimal) unit tests, 0 failed" {
set num_ran $expect_out(1,string)
gdb_assert "..."
("uplevel" body line 1)
invoked from within
"uplevel $body" NONE : spawn id exp7 not open
UNRESOLVED: gdb.server/unittest.exp: unit tests
...
The problem is (as fixed for avr in commit df5b887608 ("gdb/testsuite: better
handle failures in simavr board, reap simavr process")), that gdb_expect through
remote_expect adds a "-i <gdb spawn id> -timeout 10", which is the one causing
the error.
As in aforementioned commit, fix this by using expect instead.
Tested on x86_64-linux.
With test-case gdb.server/stop-reply-no-thread-multi.exp and host board
local-remote-host-notty.exp we occasionally run into a silent out, due to
getting:
...
(gdb) kill^M
(gdb) The program is not being run.^M
...
instead of the expected:
...
(gdb) kill^M
The program is not being run.^M
(gdb)
...
Likewise, we occasionally run into a nonsilent timeout:
...
(gdb) disconnect^M
(gdb) You can't do that when your target is `exec'^M
FAIL: gdb.server/stop-reply-no-thread.exp: to_disable=Tthread: t_nonstop=on: \
disconnect (timeout)
...
Typically, this results in the test-case taking more than two minutes to run.
The problem can be reproduced using just:
...
$ ssh -l $USER 127.0.0.1 gdb -q -ex kill
...
Note that ssh by default uses -T which disables pseudo-tty allocation (as
opposed to -t which forces pseudo-tty allocation):
...
$ ssh -l $USER 127.0.0.1 -T tty
not a tty
$ ssh -l $USER 127.0.0.1 -t tty
/dev/pts/5
Connection to 127.0.0.1 closed.
...
and according to https://stackoverflow.com/a/63241102 the behaviour we're
seeing is specific to using '-T'.
The related host board local-remote-host.exp does use '-t', and the only
difference between the two boards mentioned is whether editing is on or off.
Fix this by:
- moving the content of local-remote-host-notty.exp into
local-remote-host.exp
- consequently, extending the copyright years in local-remote-host.exp
- including local-remote-host.exp in local-remote-host-notty.exp
(making local-remote-host-notty.exp use '-t')
- adding -iex "set editing off" to GDBFLAGS in local-remote-host-notty.exp
This results in the test-case taking just 6 seconds to run.
Tested on x86_64-linux.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=29669
With test-case gdb.server/stop-reply-no-thread.exp and host board
local-remote-host.exp, I run into:
...
Breakpoint 1, ^[[33mmain^[[m () at ^[[32mstop-reply-no-thread.c^[[m:21^M
21 ^[[01;34mreturn^[[m ^[[35m0^[[m^[[31m;^[[m^M
(gdb) FAIL: gdb.server/stop-reply-no-thread.exp: to_disable=: t_nonstop=off: \
continue to main
...
The problem is that styling is enabled, and that is causing a regexp mismatch.
With native, styling is disabled in default_gdb_init by doing
'setenv TERM "dumb"', but that only has effect because the build (where we
execute runtest, and consequently the setenv) and the host (where we execute
gdb) are the same. For this host board however, gdb executes on a remote
host, and the setenv has no effect.
We could try to make some generic way to set TERM on the host, but for the
purposes of this test-case it seems sufficient to just add:
...
set GDBFLAGS "${GDBFLAGS} -iex \"set style enabled off\""
...
so let's go with that for now.
Tested on x86_64-linux.
I looked at all the spots using value_mark, and converted all the
straightforward ones to use scoped_value_mark instead.
Regression tested on x86-64 Fedora 34.
When it's impossible to read the FPCCR and XPSR, the unwinding is
unpredictable as the it's not possible to determine the correct
frame size or padding.
The only sane thing to do in this condition is to stop the unwinding.
Example session without this patch:
(gdb) bt
#0 SVC_Handler () at .../GPIO/GPIO_EXTI/Src/stm32f4xx_it.c:112
.../gdb/arm-tdep.c:3594: internal-error: arm_m_exception_cache: Assertion `safe_read_memory_unsigned_integer (FPCCR, ARM_INT_REGISTER_SIZE, byte_order, &fpccr)' failed.
A problem internal to GDB has been detected,
further debugging may prove unreliable.
----- Backtrace -----
0x5583bfb2a157 gdb_internal_backtrace_1
...
---------------------
This is a bug, please report it. For instructions, see:
<https://www.gnu.org/software/gdb/bugs/>.
Aborted (core dumped)
Example session with this patch:
(gdb) bt
#0 SVC_Handler () at .../GPIO/GPIO_EXTI/Src/stm32f4xx_it.c:112
warning: Could not fetch required FPCCR content. Further unwind is impossible.
#1 <signal handler called>
(gdb)
Reviewed-by: Pedro Alves <pedro@palves.net>
Signed-off-by: Torbjörn SVENSSON <torbjorn.svensson@foss.st.com>
Where sub and subf forms of an instruction exist we generally
disassemble to the extended insn sub form rather than the underlying
machine subf instruction. Do so for SPE evsubw and evsubiw too.
spe_ambiguous.d always was a bit too optimistic. There is no sensible
way to disassemble identical bytes back to different and original
source. Instead change the test to check -Mraw results.
gas/
* testsuite/gas/ppc/ppc.exp: Run spe_ambiguous test.
* testsuite/gas/ppc/spe.d: Expect evsubw and evsubiw rather than
evsubfw and evsubifw.
* testsuite/gas/ppc/spe_ambiguous.s: Test evnor form equivalent
to evnot.
* testsuite/gas/ppc/spe_ambiguous.d: Test Mraw.
opcodes/
* ppc-opc.c (powerpc_opcodes): Move evsubw before evsubfw and
evsubiw before evsubifw and mark EXT.
It has bothered me for a long time that we have disabled LSP (and SPE)
tests. Also the LSP test comment indicating there is something wrong
with get_powerpc_dialect. I don't think there is. Decoding of a VLE
instruction depends on whether the processor is in VLE mode (some
processors support both VLE and standard PPC) which we flag per
section with SHF_PPC_VLE for decoding when disassembling.
Background: Some versions of powerpc e200 have "Lightweight Signal
Processing" support, examples being e200z215 and e200z425. As far as
I can tell, LSP and SPE are mutually exclusive. This seems to be
borne out by insn encoding, for example LSP "zvaddih" and SPE "evaddw"
have the same encoding. So none of the processor descriptions in
ppc_opts ought to have both PPC_OPCODE_LSP and PPC_OPCODE_SPE/2, if we
want disassembly to work. I also could not find anything to suggest
that the LSP insns are enabled only in VLE mode, which means the LSP
insns should not be in vle_opcodes.
Fix all this by moving the LSP insns to their own table, and add a new
e200z2 cpu entry with LSP support, removing LSP from -me200z4 and from
-mvle. (Yes, I know, as I said above some of the e200z4 processors
have LSP. Others have SPE. It's hard to choose good options. Think
of z2 as meaning earlier, z4 as later.) Also add -mlsp to allow
adding the LSP insn set.
include/
* opcode/ppc.h (lsp_opcodes, lsp_num_opcodes): Declare.
(LSP_OP_TO_SEG): Define.
binutils/
* doc/binutils.texi: Update ppc docs.
gas/
* config/tc-ppc.c (ppc_setup_opcodes): Add lsp opcodes to ppc_hash.
* doc/c-ppc.texi: Document e200 and lsp.
* testsuite/gas/ppc/lsp-checks.d: Assemble with -me200z2.
* testsuite/gas/ppc/lsp.d: Likewise, disassembly too.
* testsuite/gas/ppc/ppc.exp: Don't xfail lsp test.
opcodes/
* ppc-dis.c (ppc_opts): Add e200z2 and lsp. Don't set
PPC_OPCODE_LSP for e200z4 or vle.
(ppc_parse_cpu): Mutually exclude LSP and SPE.
(LSP_OPCD_SEGS): Define.
(lsp_opcd_indices): New array.
(disassemble_init_powerpc): Init lsp_opcd_indices.
(lookup_lsp): New function.
(print_insn_powerpc): Call it.
* ppc-opc.c: Include libiberty.h for ARRAY_SIZE and use throughout.
(vle_opcodes): Move LSP opcodes to..
(lsp_opcodes): ..here, and sort.
(lsp_num_opcodes): New.
Besides not initialising the_bfd of synthetic symbols, counting
symbols when sizing didn't match symbols created if there were any
dynsyms named "". We don't want synthetic symbols without names
anyway, so get rid of them. Also, simplify and correct sanity checks.
PR 29677
* mach-o.c (bfd_mach_o_get_synthetic_symtab): Rewrite.
I noticed in gdb.base/skip-solib.exp:
...
if {[gdb_compile_shlib ${srcdir}/${subdir}/${srcfile_lib} ${binfile_lib} \
[list debug -Wl,-soname,${libname}.so]] != ""} {
return -1
}
...
that the -Wl,-soname argument is missing an ldflags= prefix, but adding it
gives us a duplicate:
...
Executing on host: gcc -fno-stack-protector \
outputs/gdb.base/skip-solib/skip-solib-lib.c.o -fdiagnostics-color=never \
-shared -g -Wl,-soname,libskip-solib.so -Wl,-soname,libskip-solib.so -lm \
-o outputs/gdb.base/skip-solib/libskip-solib.so (timeout = 300)
...
so apparently it's taken care of by gdb_compile_shlib.
Drop the inactive and also unnecessary -Wl,-soname,${libname}.so from the
flags list for the gdb_compile_shlib call.
Tested on x86_64-linux.
With test-case gdb.base/infoline-reloc-main-from-zero.exp and target board
unix/-fPIE/-pie I run into:
...
gdb compile failed, ld: infoline-reloc-main-from-zero: error: \
PHDR segment not covered by LOAD segment
collect2: error: ld returned 1 exit status
...
When running with native, I find that the executable is static:
...
$ file infoline-reloc-main-from-zero
infoline-reloc-main-from-zero: ELF 64-bit LSB executable, x86-64, \
version 1 (SYSV), statically linked, BuildID[sha1]=$hex, with debug_info, \
not stripped
...
despite not having been compiled with -static.
Fix the compilation by adding -static to the compilation flags.
Tested on x86_64-linux.
Compilers default to either PIE or no-PIE executables.
In order to test PIE executables with a compiler that produces non-PIE by
default, we can use target board unix/-fPIE/-pie, which set the multilib_flags
of the target board to "-fPIE -pie".
Likewise, we can use target board unix/-fno-PIE/-no-pie with a compiler that
produces PIE by default.
The target board unix/-fno-PIE/-no-pie has a potential problem when compiling
shared libs, because the multilib_flags will override the attempts of
gdb_compile_shlib to compile with -fPIC. This is taken care of by running the
body of gdb_compile_shlib wrapped in with_PIE_multilib_flags_filtered.
The target board unix/-fPIE/-pie has a problem with nopie compilations. The
current approach is to do the compilation hoping for the best, and if we find
out that the resulting executable is PIE despite specifying nopie, we error
out with the standard error message "nopie failed to prevent PIE executable".
That however does not work for hard-coded assembly nopie test-cases, which will
just noisily refuse to compile:
...
ld: amd64-disp-step0.o: relocation R_X86_64_32S against `.text' can not be \
used when making a PIE object; recompile with -fPIE^M
...
Fix this in gdb_compile by filtering out the PIE settings in the target board
multilib_flags when pie or nopie is specified.
Tested on x86_64-linux.
Add a new proc cond_wrap, that can be used to replace the repetitive:
...
if { $cond } {
wrap {
<body>
}
} else {
<body>
}
...
with the shorter:
...
cond_wrap $cond wrap {
<body>
}
...
Tested on x86_64-linux.
The lack thereof did caused gas to issue "internal: unreachable
INSN_CLASS_*" errors when trying to assemble respective insns without
the feature(s) enabled via e.g. ".option arch, ...". Of course a proper
hint towards the missing extension then wasn't given either.
'H', 'Smstateen', 'Sscofpmf' and 'Sstc' are four privileged extensions with
their CSR definitions and 'Smepmp' is a privileged extension with additional
CSR bits.
Volume II: Privileged Architecture of the RISC-V ISA Manual states that the
privileged architecture requires the 'Zicsr' extension. However, current
GNU Binutils has no direct way whether the program has dependency to the
privileged architecture itself.
As a workaround, we should add implications from privileged extensions that
either add new CSRs, extend existing CSRs or depends on using CSRs.
This commit adds such implications for existing privileged extensions that
satisfy this condition.
gas/ChangeLog:
* testsuite/gas/riscv/march-imply-h.d: New test, at least for 'H'.
bfd/ChangeLog:
* elfxx-riscv.c (riscv_implicit_subsets): Add 'Zicsr'
implicications for privileged extensions 'H', 'Smstateen',
'Sscofpmf', 'Sstc' and 'Smepmp'.
Before changing the core disassembler, we take care of minor code clarity
issues and improve readability.
This commit removes unused variable last_map_state (set by the
print_insn_riscv function but not read anywhere else).
opcodes/ChangeLog:
* riscv-dis.c (last_map_state): Remove.
(print_insn_riscv): Remove setting last_map_state.
Before changing the core disassembler, we take care of minor code clarity
issues and improve readability.
Since xlen variable is not (and should not) used outside riscv-dis.c,
this commit makes this variable static.
opcodes/ChangeLog:
* riscv-dis.c (xlen): Make this variable static.
Before changing the core disassembler, we take care of minor code clarity
issues and improve readability.
This commit replaces uses of int with bool whenever possible.
opcodes/ChangeLog:
* riscv-dis.c (no_aliases) Change type to bool.
(set_default_riscv_dis_options): Use boolean.
(parse_riscv_dis_option_without_args): Likewise.
(riscv_disassemble_insn): Use boolean keywords.
Before changing the core disassembler, we take care of minor code clarity
issues and improve readability.
This commit takes care of improper spacing for code clarity.
opcodes/ChangeLog:
* riscv-dis.c (riscv_disassemble_insn): Tidying with spacing.
Before changing the core disassembler, we take care of minor code clarity
issues and improve readability.
First, we need to clarify the roles of variables and code portions.
opcodes/ChangeLog:
* riscv-dis.c (xlen): Move before default_isa_spec. Add comment.
(default_isa_spec, default_priv_spec): Add comment.
(riscv_gpr_names, riscv_fpr_names): Likewise.
(parse_riscv_dis_option_without_args): Likewise.
(parse_riscv_dis_option, parse_riscv_dis_options): Likewise.
(maybe_print_address): Likewise.
(riscv_disassemble_insn): Fix comment about the Zfinx "extension".
Add comment about the riscv_multi_subset_supports call.
Because all standard hints must be placed before corresponding instruction
for the disassembler, they may taint basic RVI instruction section.
This commit moves all standard hints before all basic RVI instructions
to improve maintainability.
opcodes/ChangeLog:
* riscv-opc.c (riscv_opcodes): Move all standard hints before all
standard instructions.
This is a part of small tidying (declare tables in riscv-opc.c).
include/ChangeLog:
* opcode/riscv.h (riscv_rm, riscv_pred_succ): Move declarations to
opcodes/riscv-opc.c. New non-static definitions.
opcodes/ChangeLog:
* riscv-opc.c (riscv_rm, riscv_pred_succ): Move from
include/opcode/riscv.h. Add description.
Test gdb.base/watchpoint.exp generates 4 test errors on Power 9. The
test uses the test [target_info exists gdb,no_hardware_watchpoints] to
determine if the processor supports hardware watchpoints. The check
only examines the processor type to determine if it supports hardware
watchpoints.
The PowerPC processors support hardware watchpoints with the
exception of Power 9. The hardware watchpoint support is disabled on
Power 9. The test skip_hw_watchpoint_tests must be used to correctly
determine if the PowerPC processor supports hardware watchpoints.
This patch replaces the [target_info exists gdb,no_hardware_watchpoints]
with the skip_hw_watchpoint_tests_p check. With the patch, the test runs
on Power 9 with hardware watchpoint force-disabled. The test runs on
all other PowerPC processors with and without hardware watchpoints
enabled.
The patch has been tested on Power 9 to verify the test only runs with
hardware breakpoints disabled. The patch has been tested on X86-64 with
no regression failures. The test fails on Power 10 due to an internal GDB
error due to resource management. The resource management issue will be
addressed in another patch.
With test-case gdb.dwarf2/macro-source-path.exp and target board unix/-m32, I
run into:
...
as: macro-source-path-gcc11-ld238-dw5-filename-641.o: \
unsupported relocation type: 0x1^M
...
The problem is that we have 64-bit dwarf so the debug_line offset in the
.debug_macro section is an 8-byte entity, emitted using ".8byte":
...
.section .debug_macro
.Lcu_macros4:
.2byte 5 /* version */
.byte 3 /* flags */
.8byte .LLlines3 /* debug_line offset */
...
but the linker doesn't support 8-byte relocation types on a 32-bit architecture.
This is similar to what was fixed in commit a5ac8e7fa3
("[gdb/testsuite] Fix 64-bit dwarf test-cases with -m32") for for instance
.debug_abbrev.
Fix this in the same way, by using _op_offset to emit the debug_line offset.
Tested on x86_64-linux with native and target board unix/-m32.
With test-case gdb.dwarf2/entry-value-typedef.exp and target board unix/-m32,
I run into:
...
builtin_spawn -ignore SIGHUP g++ -fno-stack-protector \
gdb/testsuite/gdb.dwarf2/entry-value-typedef-amd64.S \
-fdiagnostics-color=never -Lbuild/libiberty -lm -m32 \
-o outputs/gdb.dwarf2/entry-value-typedef/entry-value-typedef^M
entry-value-typedef.cpp: Assembler messages:^M
entry-value-typedef.cpp:38: Error: bad register name `%rbp'^M
...
The problem is that the test-cases selects an amd64 .S file based on the check:
...
if { [istarget "x86_64-*-linux*"] } {
...
which is also true for target board unix/-m32 on x86_64-linux.
Fix this by adding the missing is_lp64_target check.
Tested on x86_64-linux, using native and target board unix/-m32.
With target board unix/-m32 and test-case gdb.mi/mi-disassemble.exp we have:
...
(gdb) ^M
print/x *((unsigned char *) 0x8048485)^M
&"print/x *((unsigned char *) 0x8048485)\n"^M
~"$9 = 0x83\n"^M
^done^M
(gdb) ^M
PASS: gdb.mi/mi-disassemble.exp: get valueof "*((unsigned char *) 0x8048485)"
FAIL: gdb.mi/mi-disassemble.exp: byte at 0x8048485 matches
...
The test-case passes with native.
With native we see in gdb.log that variable longest_insn_bytes is:
...
Longest instruction at 0x0000000000400549 with bytes '48 8b 05 20 01 00 00'
...
and variable split_bytes (added debug puts) ends up as:
...
SPLIT_BYTES: 48 8b 05 20 01 00 00
...
But with unix/-m32 we have longest_insn_byte:
...
Longest instruction at 0x08048481 with bytes '8d 4c 24 04 '
...
and split_bytes ends up as:
...
SPLIT_BYTES: 8d 4c 24 04 {} {} {} {} {} {} {} {}
...
so the trailing whitespace is translated by split to empty bytes, and the
mismatch FAILs are generated for those.
Fix this by stripping the whitespace, which makes us end up with a different
and indeed longer insn:
...
Longest instruction at 0x08048492 with bytes 'dd 05 98 85 04 08'
...
Tested on x86_64-linux, with native and target board unix/-m32.
Test gdb.base/watchpoint-stops-at-right-insn.exp generates 4 test errors
on Power 9. The test uses the test [target_info exists gdb,
no_hardware_watchpoints] to determine if the processor supports hardware
watchpoints. The check only examines the processor type to determine if
it supports hardware watchpoints. Note, the test works fine on Power 10.
The PowerPC processors support hardware watchpoints with the
exception of Power 9. The hardware watchpoint support is disabled on
Power 9. The test skip_hw_watchpoint_tests must be used to correctly
determine if the PowerPC processor supports hardware watchpoints.
This patch replaces the [target_info exists gdb,no_hardware_watchpoints]
with the skip_hw_watchpoint_tests_p check. With the patch, the test is
disabled on Power 9 but runs on all other PowerPC processors.
The patch has been tested on Power 9, Power 10 and X86-64 with no
regression failures.
When running test-case gdb.base/ctf-constvars.exp on openSUSE Tumbleweed (with
system gcc version 12, providing gcc -gctf support, enabling the ctf test-cases
in the gdb testsuite), I run into:
...
(gdb) print vox^M
'vox' has unknown type; cast it to its declared type^M
(gdb) FAIL: gdb.base/ctf-constvars.exp: print vox
...
There are two causes for this:
- the linker flags are missing --ctf-variables, so the information for variable
vox is missing (reported in PR29468), and
- the executable contains some dwarf2 due to some linked-in glibc objects,
so the ctf info is ignored (reported in PR29160).
By using:
- -Wl,--ctf-variable,
- -Wl,--strip-debug, and
we can make the test-case and some similar test-cases pass.
Tested on x86_64-linux.
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=29160
Bug: https://sourceware.org/bugzilla/show_bug.cgi?id=29468
When running test-case gdb.base/gdbindex-stabs.exp on openSUSE Tumbleweed (with
gcc 12) I get:
...
gdb compile failed, gdb/testsuite/gdb.base/gdbindex-stabs.c: warning: \
STABS debugging information is obsolete and not supported anymore
...
Silence the warning by passing quiet to gdb_compile. Likewise in two other
test-cases.
With test-cases gdb.base/cvexpr.exp and gdb.base/whatis.exp I run into:
...
gdb compile failed, gcc: error: unrecognized debug output level 't'
...
This is due to using additional_flags=-gt.
Commit ffb3f58793 ("CTF: multi-CU and archive support") replaced
additional_flags=-gt with additional_flags=-gctf in gdb.ctf/*.exp and
gdb.base/ctf-*.exp.
Do the same in these two test-cases.
Tested on x86_64-linux.
On openSUSE Tumbleweed (with ld 2.39) and test-case
gdb.base/infoline-reloc-main-from-zero.exp, I get:
...
gdb compile failed, ld: warning: infoline-reloc-main-from-zero has a LOAD \
segment with RWX permissions
UNTESTED: gdb.base/infoline-reloc-main-from-zero.exp: \
infoline-reloc-main-from-zero.exp
...
Fix this by compiling with -Wl,--no-warn-rwx-segments.
Tested on x86_64-linux.
On openSUSE Tumbleweed (with ld 2.39) I get for test-case
gdb.base/nested-subp2.exp:
...
gdb compile failed, ld: warning: tmp.o: requires executable stack \
(because the .note.GNU-stack section is executable)
...
Fix this by compiling with -Wl,--no-warn-execstack.
Likewise in gdb.base/nested-subp3.exp
Tested on x86_64-linux.
On openSUSE Tumbleweed I noticed:
...
UNTESTED: gdb.dwarf2/fission-absolute-dwo.exp: fission-absolute-dwo.exp
ERROR: failed to compile fission-absolute-dwo
...
The ERROR is unnecessary, given that an UNTESTED is already emitted.
Furthermore, it could be argued that it is incorrect because it's not a
testsuite error to not be able to compile something, and UNTESTED or
UNSUPPORTED is more appropriate.
Remove the perror call, likewise in fission-relative-dwo.exp.
Tested on x86_64-linux.
Because "%p" is the pointer conversion specifier to print a pointer in an
implementation-defined manner, the result with format string containing
"0x%p" can be strange. For instance, core_map_find_mapping prints error
containing "0x0x...." (processor is not NULL) or "0x(null)" (processor is
NULL) on glibc.
This commit replaces "0x%p" with "%p" to prevent unpredictable behavior.
After the recent series of fixes to mark more functions in the
simulator with ATTRIBUTE_PRINTF, there were some build failures in the
ppc sim due, in some cases, to bugs with the arguments being passed,
and in other cases, the issues were (maybe) less serious, with
arguments being the wrong size, or type, for the printf format being
used.
This commit fixes all of the issues that I ran into.
In each case I selected the easiest solution to the problem, which is
usually just casting the argument to the correct type. If anyone
later on thinks the print format should change, please feel free to do
that. What we have here should keep the simulator basically working
as it does currently, which is my goal with this commit.
I noticed that the $want_gnu_debuglink code in gdb/contrib/cc-with-tweaks.sh
uses objcopy instead of $OBJCOPY. Fix this.
Script checked with shellcheck, no new warnings added.
Tested on x86_64-linux.
Commit 228cf97dd3 ("Merge configure.ac from gcc project") undid the
change originally done in de83289ef3 ("Pass PKG_CONFIG_PATH down from
top-level Makefile"). Re-apply it.
Change-Id: I91138dfca41c43b05e53e445f62e4b27882536bf
Having two overloads of target_read_auxv that don't have the same goals
is confusing. Rename the one that reads from an explicit target_ops to
target_read_auxv_raw. Also, it occured to me that the non-raw version
could use the raw version, that reduces duplication a bit.
Change-Id: I28e5f7cecbfcacd0174d4686efb3e4a23b4ad491
There's a flaw in the interaction of the auxv caching and the fact that
target_auxv_search allows reading auxv from an arbitrary target_ops
(passed in as a parameter). This has consequences as explained in this
thread:
https://inbox.sourceware.org/gdb-patches/20220719144542.1478037-1-luis.machado@arm.com/
In summary, when loading an AArch64 core file with MTE support by
passing the executable and core file names directly to GDB, we see the
MTE info:
$ ./gdb -nx --data-directory=data-directory -q aarch64-mte-gcore aarch64-mte-gcore.core
...
Program terminated with signal SIGSEGV, Segmentation fault
Memory tag violation while accessing address 0x0000ffff8ef5e000
Allocation tag 0x1
Logical tag 0x0.
#0 0x0000aaaade3d0b4c in ?? ()
(gdb)
But if we do it as two separate commands (file and core) we don't:
$ ./gdb -nx --data-directory=data-directory -q -ex "file aarch64-mte-gcore" -ex "core aarch64-mte-gcore.core"
...
Program terminated with signal SIGSEGV, Segmentation fault.
#0 0x0000aaaade3d0b4c in ?? ()
(gdb)
The problem with the latter is that auxv data gets improperly cached
between the two commands. When executing the file command, auxv gets
first queried here, when loading the executable:
#0 target_auxv_search (ops=0x55555b842400 <exec_ops>, match=0x9, valp=0x7fffffffc5d0) at /home/simark/src/binutils-gdb/gdb/auxv.c:383
#1 0x0000555557e576f2 in svr4_exec_displacement (displacementp=0x7fffffffc8c0) at /home/simark/src/binutils-gdb/gdb/solib-svr4.c:2482
#2 0x0000555557e594d1 in svr4_relocate_main_executable () at /home/simark/src/binutils-gdb/gdb/solib-svr4.c:2878
#3 0x0000555557e5989e in svr4_solib_create_inferior_hook (from_tty=1) at /home/simark/src/binutils-gdb/gdb/solib-svr4.c:2933
#4 0x0000555557e6e49f in solib_create_inferior_hook (from_tty=1) at /home/simark/src/binutils-gdb/gdb/solib.c:1253
#5 0x0000555557f33e29 in symbol_file_command (args=0x7fffffffe01c "aarch64-mte-gcore", from_tty=1) at /home/simark/src/binutils-gdb/gdb/symfile.c:1655
#6 0x00005555573319c3 in file_command (arg=0x7fffffffe01c "aarch64-mte-gcore", from_tty=1) at /home/simark/src/binutils-gdb/gdb/exec.c:555
#7 0x0000555556e47185 in do_simple_func (args=0x7fffffffe01c "aarch64-mte-gcore", from_tty=1, c=0x612000047740) at /home/simark/src/binutils-gdb/gdb/cli/cli-decode.c:95
#8 0x0000555556e551c9 in cmd_func (cmd=0x612000047740, args=0x7fffffffe01c "aarch64-mte-gcore", from_tty=1) at /home/simark/src/binutils-gdb/gdb/cli/cli-decode.c:2543
#9 0x00005555580e63fd in execute_command (p=0x7fffffffe02c "e", from_tty=1) at /home/simark/src/binutils-gdb/gdb/top.c:692
#10 0x0000555557771913 in catch_command_errors (command=0x5555580e55ad <execute_command(char const*, int)>, arg=0x7fffffffe017 "file aarch64-mte-gcore", from_tty=1, do_bp_actions=true) at /home/simark/src/binutils-gdb/gdb/main.c:513
#11 0x0000555557771fba in execute_cmdargs (cmdarg_vec=0x7fffffffd570, file_type=CMDARG_FILE, cmd_type=CMDARG_COMMAND, ret=0x7fffffffd230) at /home/simark/src/binutils-gdb/gdb/main.c:608
#12 0x00005555577755ac in captured_main_1 (context=0x7fffffffda10) at /home/simark/src/binutils-gdb/gdb/main.c:1299
#13 0x0000555557775c2d in captured_main (data=0x7fffffffda10) at /home/simark/src/binutils-gdb/gdb/main.c:1320
#14 0x0000555557775cc2 in gdb_main (args=0x7fffffffda10) at /home/simark/src/binutils-gdb/gdb/main.c:1345
#15 0x00005555568bdcbe in main (argc=10, argv=0x7fffffffdba8) at /home/simark/src/binutils-gdb/gdb/gdb.c:32
Here, target_auxv_search is called on the inferior's target stack. The
target stack only contains the exec target, so the query returns empty
auxv data. This gets cached for that inferior in `auxv_inferior_data`.
In its constructor (before it is pushed to the inferior's target stack),
the core_target needs to identify the right target description from the
core, and for that asks the gdbarch to read a target description from
the core file. Because some implementations of
gdbarch_core_read_description (such as AArch64's) need to read auxv data
from the core in order to determine the right target description, the
core_target passes a pointer to itself, allowing implementations to call
target_auxv_search it. However, because we have previously cached
(empty) auxv data for that inferior, target_auxv_search searched that
cached (empty) auxv data, not auxv data read from the core. Remember
that this data was obtained by reading auxv on the inferior's target
stack, which only contained an exec target.
The problem I see is that while target_auxv_search offers the
flexibility of reading from an arbitrary (passed as an argument) target,
the caching doesn't do the distinction of which target is being queried,
and where the cached data came from. So, you could read auxv from a
target A, it gets cached, then you try to read auxv from a target B, and
it returns the cached data from target A. That sounds wrong. In our
case, we expect to read different auxv data from the core target than
what we have read from the target stack earlier, so it doesn't make
sense to hit the cache in this case.
To fix this, I propose splitting the code paths that read auxv data from
an inferior's target stack and those that read from a passed-in target.
The code path that reads from the target stack will keep caching,
whereas the one that reads from a passed-in target won't. And since,
searching in auxv data is independent from where this data came from,
split the "read" part from the "search" part.
From what I understand, auxv caching was introduced mostly to reduce
latency on remote connections, when doing many queries. With the change
I propose, only the queries done while constructing the core_target
end up not using cached auxv data. This is fine, because there are just
a handful of queries max, done at this point, and reading core files is
local.
The changes to auxv functions are:
- Introduce 2 target_read_auxv functions. One reads from an explicit
target_ops and doesn't do caching (to be used in
gdbarch_core_read_description context). The other takes no argument,
reads from the current inferior's target stack (it looks just like a
standard target function wrapper) and does caching.
The first target_read_auxv actually replaces get_auxv_inferior_data,
since it became a trivial wrapper around it.
- Change the existing target_auxv_search to not read auxv data from the
target, but to accept it as a parameter (a gdb::byte_vector). This
function doesn't care where the data came from, it just searches in
it. It still needs to take a target_ops and gdbarch to know how to
parse auxv entries.
- Add a convenience target_auxv_search overload that reads auxv
data from the inferior's target stack and searches in it. This
overload is useful to replace the exist target_auxv_search calls that
passed the `current_inferior ()->top_target ()` target and keep the
call sites short.
- Modify parse_auxv to accept a target_ops and gdbarch to use for
parsing entries. Not strictly related to the rest of this change,
but it seems like a good change in the context.
Changes in architecture-specific files (tdep and nat):
- In linux-tdep, linux_get_hwcap and linux_get_hwcap2 get split in two,
similar to target_auxv_search. One version receives auxv data,
target and arch as parameters. The other gets everything from the
current inferior. The latter is for convenience, to avoid making
call sites too ugly.
- Call sites of linux_get_hwcap and linux_get_hwcap2 are adjusted to
use either of the new versions. The call sites in
gdbarch_core_read_description context explicitly read auxv data from
the passed-in target and call the linux_get_hwcap{,2} function with
parameters. Other call sites use the versions without parameters.
- Same idea for arm_fbsd_read_description_auxv.
- Call sites of target_auxv_search that passed
`current_inferior ()->top_target ()` are changed to use the
target_auxv_search overload that works in the current inferior.
Reviewed-By: John Baldwin <jhb@FreeBSD.org>
Reviewed-By: Luis Machado <luis.machado@arm.com>
Change-Id: Ib775a220cf1e76443fb7da2fdff8fc631128fe66
